Abstract: Inverter is widely used in UPS, motors, and renewable energy generation systems. Its safe operation is of great significance for converter systems. In harsh working environments, the power transistor and current sensor of grid connected inverter are prone to faults, and their fault characteristics are similar and coupled, posing great challenges to existing inverter fault diagnosis technologies. Therefore, this paper proposes a coupling fault diagnosis method based on conditional generative adversarial network（CGAN）and long short-term memory（LSTM）for inverters. First, the basic working principle of the inverter is analyzed and a digital model is established on the dSPACE platform. Next, the working modes under single power transistor open circuit fault, double power transistor open circuit faults and zero output fault of the current sensor are explored through the model. Then, three phase currents are used as diagnosis variables, and CGAN is utilized to obtain the fault data which is closed to the real operation conditions. The faults of switches and sensors are finally located by the LSTM network. The experimental data proves the effectiveness of this method.